Home > Storage > PowerScale (Isilon) > Industry Solutions and Verticals > Electronic Design Automation > EDA Cloud Burst with Dell PowerScale and Vcinity Data Access > Test results and findings
Baseline results on on-premises environment are used for comparing on-premises performance to that of cloud bursting. Submit 12, 24, and 48 jobs using job scheduler, and compare to Vcinity Data Access Appliance with and without cache results.
Vcinity ULT X creates a highly efficient RDMA connection between on-premises and cloud. For this test, cache is not used on the Vcinity appliance at the destination to test native RDMA over WAN performance. Millions of small files are transferred across the WAN for Cloud Burst.
Each android build job will pull main branch source code from the on-premises NFS share for job execution. The source code contains millions of small files like EDA workspace.
Emulate WAN latency of 30ms and 60ms using WAN emulator. The higher the latency, the longer it will take to transfer all the data required for the EDA workload for Cloud Burst. Latency over 60ms is not recommended for a Cloud Burst type of workload. The testing proved that beyond 60ms latency, the EDA jobs began to timeout.
Vcinity ULT X creates a highly efficient RDMA connection between on-premises and cloud. For this test, cache is used on the Vcinity system at the destination, thus allowing the EDA dataset to be accessed as if it were a local copy. This cache acts primarily as a read cache for this workload but can also perform as an accelerated write-back cache when data and results need to be committed back to the source location.
When the file is read for the first time, the data will be transferred to the cloud through the highly efficient RDMA connection and cached on the Vcinity appliance. Initial sync will transfer only the metadata, files, and directories the EDA job requests. Next time the file is read, the data is available immediately from the cache.
In the case of Cloud Burst EDA workloads, libraries and tools directories are kept in cache because these directories are accessed very frequently during EDA workloads. The design team accesses the same project workspaces for EDA workloads, so there is a high chance all data is available immediately from the cache.
The chart below shows the effectiveness of utilizing the Vcinity cache capability. After the initial run, most of the EDA dataset is cached on the remote Vcinity appliance. During subsequent runs, data is accessed immediately from the cache. The total elapsed time of 12 and 24 jobs is almost the same as baseline results on 30ms and 60ms with little performance degradation on 120ms. Testing proved that once the EDA dataset was hosted in the Vcinity cache at the destination (public cloud), performance was similar to that of on-premises.